Taiwan Artificial Intelligence Basic Act

What it is

The Artificial Intelligence Basic Act is Taiwan’s framework law for AI. The Legislative Yuan passed it on third reading on 23 December 2025, and it was promulgated on 14 January 2026. Under Article 20 it took effect on promulgation. The Act has 20 articles. The Laws and Regulations Database publishes an English translation, and the Chinese text prevails where they differ.

It is a basic act in the Taiwanese sense. It sets the direction for government policy and allocates responsibilities, and it leaves detailed obligations to later regulation. Its stated purposes are building a smart nation, promoting human-centric AI, safe AI applications, digital equity and ethical use of technology.

The National Science and Technology Council (NSTC) is the central competent authority. A National AI Strategic Committee under the Executive Yuan, chaired by the Premier, coordinates national AI policy. The Ministry of Digital Affairs (MODA) leads on risk.

Who it applies to

The Act’s duties are addressed to government at central and local level. It defines AI as an autonomous system that uses machine learning and algorithms to produce outputs such as predictions, content, recommendations or decisions that affect physical or virtual environments.

Businesses that develop, provide or use AI in Taiwan are affected in three ways. Sector regulators must write risk-based rules and industry guidelines for AI in their fields. Applications designated high risk carry warning and notice duties. And government must clarify liability for high-risk AI and set up relief, compensation or insurance mechanisms. Pre-deployment research and development is excluded from that liability provision unless it is tested in real-world settings or used to provide products or services.

What it requires

• Seven principles. Article 4 requires government AI work to follow sustainable development and well-being, human autonomy, privacy protection and data governance, cybersecurity and safety, transparency and explainability, fairness and non-discrimination, and accountability.
• Harm prevention. Article 5 requires government to prevent AI applications from harming life, body, liberty or property, disrupting social order, national security or the environment, or breaching law through bias, false advertising or misleading information. Such uses must be restricted or prohibited by law under Article 16.
• High-risk designation. Competent authorities, in consultation with MODA, may designate high-risk applications, which must then carry advisory notices or warnings. MODA and other agencies provide assessment tools.
• Risk framework. Article 16 directs MODA to promote an AI risk classification and assessment framework aligned with international standards, and sector regulators to build risk-based rules on it.
• Personal data. Article 14 requires authorities to work with the data protection authority and calls for privacy by design and by default, avoiding unnecessary processing of personal data.
• Government use. Article 19 requires agencies that use AI in their work or services to carry out risk assessments and set usage guidelines and internal controls.

MODA’s AI Risk Classification Framework, version 1.0 dated 7 July 2026, is the first instrument under Article 16. It gives agencies a common vocabulary and four steps. Agencies take stock of AI applications in their sector, identify the risk types that arise, assess how much each could affect people or systems, and respond with management measures. It covers civilian uses and excludes military AI.

Key dates

The Act passed on 23 December 2025 and took effect on 14 January 2026. MODA’s risk framework v1.0 followed in July 2026. Article 18 gives agencies two years from commencement, to January 2028, to bring their laws and administrative measures into line.

How it relates to other frameworks

Taiwan’s approach sits closest to Japan’s AI Promotion Act, a principles-based law built around government policy rather than detailed duties and penalties for business. It is lighter than South Korea’s AI Basic Act and Vietnam’s AI law, both of which place obligations on providers of high-risk AI directly. For organisations preparing now, the MODA framework’s four steps map onto the risk identification and treatment practices in the NIST AI RMF and ISO/IEC 42001, so work already done against those carries across as sector rules arrive.

A note on this page

This is Aicura’s reading of the Act, written to help you understand it. It is not legal advice and it is not the Act itself. Read the primary sources on the Laws and Regulations Database and from MODA, and take advice on how they bear on your organisation.